基于人工神经网络响应面技术的多学科优化方法研究及应用

Research and application of multidisciplinary design optimization based on artificial neural network response surface

在常用的非层次型多学科综合优化算法基础上,提出了基于神经网络响应面多学科优化算法(ANNMDO),是一种二级结构优化方法,子学科在优化时只需满足本学科的局部约束,学科层优化目标是使该学科优化设计方案与系统层提供的目标方案差异最小,系统层提供一种协调各个学科优化结果冲突机制,并且所需学科层信息通过神经网络响应面获取.最后将协同优化算法(CO)、并行子空间优化算法(CSSO)、ANNMDO算法应用于齿轮减速箱算例,验证了本文算法的高效性.

On the basis of non-hierarchical multidisciplinary optimization algorithm,this paper proposed a new multidisciplinary design optimization based on the neural network response surface（ANN MDO）,a two-level optimization architecture.That is to say,the sub-discipline level only meet the local constraints and the objective is to get smallest difference between local optimal solution and target program provided by system level.Meanwhile,the system level not only offers some coordinating mechanism to guarantee agreement of all discipline level optimal solution,but also obtains discipline level information by artificial neural network-based response surface approximation.Finally,a gearbox is adopted as an example to verify the efficiency of ANN MDO algorithm,which compare the collaborative optimization（CO） with concurrent subspace optimization（CSSO）.